Press "Enter" to skip to content

Posts published in “News”

Anthropic Establishes Fifth APAC Hub in Singapore, Signing AI Safety Pact

October 14, 2026

Anthropic;

Today, we are expanding our Asia-Pacific operations by opening a new office in Singapore—our fifth in the region—and signing a Memorandum of Understanding (MOU) with the Singapore Ministry of Digital Development and Information. Singapore is one of our most active markets anywhere in the world, and Claude is already powering Singaporean business and government services used by millions of people every day, like GovTech Singapore, Grab, and DBS. Our growing local team will ensure these organizations and many more stay on the frontier with safe, capable AI.

The AI Safety Pact is a non-binding framework established with the government to advance AI safety, testing, and responsible use. Building on prior multilingual red-teaming work, it drives joint action on security, incident response, evaluation standards, and economic research.

Google is officially rolling out Gemini 4 Argon

October 14, 2026

Gemini 4 Argon is Google DeepMind’s flagship AI model, engineered for complex, long-horizon tasks across real-world software engineering, enterprise workflows, and cybersecurity defense.

Featuring a 1-million-token output limit, Argon can reason continuously to solve difficult multi-step problems in a single pass. It excels at large-scale codebase migrations, data center memory optimization, financial analysis, and vulnerability patching. It also features hardened defenses against indirect prompt injection.

Following an initial phased rollout to trusted security partners, Argon is accessible to developers and businesses via the Gemini API and Google Cloud.

From blog.google;

Argon excels at deep work in complex, real-world tasks across domains including coding, finance, legal, cybersecurity, and science. Its depth and precision across multi-step tasks leads Argon to outperform other frontier models on benchmarks across domains, ranking #1 on the Vals Index, which measures model performance across critical economic sectors like finance, coding, legal, and tax work.

The model launches today at an introductory price of $2 per million input tokens and $10 per million output tokens (with cached input tokens at 95% off).

Anthropic warns AI models may soon self-replicate and could wreak havoc on the web  

October 14, 2026

Raphael Satter and Harshita Mary Varghese, reporting for Reuters;

A new warning about artificial intelligence from Anthropic caused a stir in cybersecurity circles on Tuesday, a ​day after the company warned that AI models could be months away from self-replicating across the internet.

Not that all surprising, specially if AI company like Anthropic, OpenAI and more, include in their AI models that ability to “self-improvement” and “self-sustaining”.

Here’s the intro from Anthropic’s article – AI models may be capable of self-replicating without access to a datacenter.

Our experiments suggest that small AI models may be months away from being capable enough to self-replicate on commercial internet-of-things (IoT) devices, like smart ovens and off-brand Android TV boxes. A replicator like this would be challenging to shut down, since there’s no centralized point to shut off—though it may be slow (with a doubling time of days). Unlike non-AI malware, it could also figure out unfamiliar devices on its own. In our evaluations, a 9B model got inference running on 8 of the 9 hardware types we tested.

Models of 30B parameters and under can already complete many of the individual steps for replication today, though they make mistakes when composing them end to end. In our simplified evaluations, GLM-5.2 and stronger models are capable of end to end replication. If small models improve at their historical rate, they’ll match this capability level by the end of this year. Small weight-based replicators may be possible soon after.

What caught my eye is that Anthropic liken this AI behavior to a “computer worm” or a malware.. specifically an “AI-malware” which very concerning.

Google Limits Free Gemini Tier to Flash-Lite in New Update

October 6, 2026

Google recently updated its official support documentation regarding changes to Gemini model access and usage limits, formalizing a significant shift in how its AI tiers are structured:

According to the updated support page, Google is restricting model availability for personal accounts starting October 9. Free users will now be limited strictly to the lightweight Flash-Lite model for lightweight tasks like email rewrites or summaries, losing access to both Flash and Pro models. Meanwhile, AI Plus subscribers ($4.99/month) retain Flash-Lite and Flash but lose Pro tier access. Higher-tier subscribers—Google AI Pro and AI Ultra—maintain full model availability along with specialized features like Deep Think for maximum parallel reasoning. Google notes that rollout dates for paying subscribers will arrive via email.

Here’s a table on the different Gemini Model availability this October 9th.

Gemini Model
Google AI Plan Flash-Lite Flash Pro
Without a plan Checkmark
AI Plus Checkmark Checkmark
AI Pro Checkmark Checkmark Checkmark
AI Ultra Checkmark Checkmark Checkmark

Simultaneously, Google is rolling out free “Skills” tools that allow users to create reusable, stackable workflows directly inside chats for writing, research, or business tasks—effectively replacing the older Gems feature. The shift has sparked notable frustration among free users, who view the model restrictions as a severe downgrade and are openly weighing alternatives like ChatGPT.

Google Unveils Gemini 4 Argon, but Access Is Limited

October 5, 2026

Google announced Gemini 4 Argon on September 30, a frontier model built to sustain deep reasoning across complex, long-horizon workflows in software engineering, finance, law, and cybersecurity defense.

Argon is not yet broadly available. It is rolling out first to trusted cyber defenders through the Fairwind Program, while Google participates in the U.S. government’s voluntary pre-release model access process. Paid API customers and Google AI Ultra subscribers will be first in line for the general release, though Google has not provided a specific date.

Argon launches with introductory pricing of $2 per million input tokens and $10 per million output tokens (discounted by 95% for cached inputs), which will eventually rise to $4 and $20. Google also expanded the model’s maximum output limit from 64,000 tokens to 1 million. According to independent firm Artificial Analysis, Argon matches GPT-6 Astra on its Intelligence Index at 60% of the cost per task at these introductory rates.

Google is already deploying Argon internally. In quantum computing research, the model beat a published optimization baseline by 40%, and in infrastructure, it identified memory optimizations capable of saving up to 1 PiB across Google data centers. Argon agents are also migrating legacy C/C++ code to Rust: in the libgav1 video decoder, they replaced 32,000 lines of SIMD code with Rust that executes 2.7 times faster than previous manual ports.

On security, Argon can autonomously find, validate, and patch critical vulnerabilities—including a previously undiscovered flaw in healthcare software that exposed patient data across global hospitals. Google also plans to offer a dedicated version without cyber guardrails specifically for trusted defenders and internal safety teams.

As industry momentum shifts toward personal agents, Google’s product chief confirmed the company is exploring integrating Argon to power complex, multi-step tasks within its Spark platform.

Anthropic Launches Claude Sonnet 5.5, a Faster, Lower-Cost Sonnet

September 29, 2026

Anthropic has introduced Claude Sonnet 5.5, the second model in the Claude 5.5 family. It is a clear upgrade over Sonnet 5, running more than 30% faster and reducing overall execution costs by up to 30% for most tasks.

Sonnet 5.5 is designed for well-scoped everyday tasks, bug fixes, and polished documents, slides, and spreadsheets. It serves as a faster, lower-cost complement to Opus 5.5, which is built for complex work requiring careful judgment. Token pricing remains unchanged from Sonnet 5 at $2 per million input tokens and $10 per million output tokens.

On benchmarks, Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0—an agentic coding benchmark—compared to 10.3% for Sonnet 5 and 66.4% reported for Opus 5.5 on the same test. It is also the first Sonnet model to beat Pokémon Red operating solely from screenshots.

Because its cybersecurity capabilities are comparable to Opus 5, it is the first Sonnet model to launch with cyber safeguards and fallbacks similar to those on Anthropic’s flagship models. Its biological safety safeguards remain unchanged.

Sonnet 5.5 is available on AWS, Google Cloud, Microsoft Azure, and the Claude Platform as claude-sonnet-5-5. Developers running Sonnet with thinking disabled must switch to the new between_tools setting prior to deployment. Claude Haiku 5.5 will join the family in the coming weeks.

WSJ: OpenAI Scraps Release of New AI Model Over Safety Concerns  

September 29, 2026

Maxwell Zeff, reporting for Wall Street Journal;

OpenAI says it is scrapping the release of its next-generation AI model over safety concerns that researchers raised during internal testing, in one of the clearest signs so far that agent misbehavior could stymie the industry’s rapid progression.

The move follows a summer punctuated by reports of artificial-intelligence systems industrywide going rogue, and marks a rare case of a major AI developer ditching a new release because of safety concerns.

This announcement comes less than a month after unveiling GPT-6 Astra and calling it “the world’s most intelligent and aligned model.“

ICYMI: Meta announces Muse: The World’s First Personal AI Agent Built for Everyone

September 28, 2026

From Meta;

Today, Meta is introducing Muse, a secure, private personal AI agent that proactively helps with people’s goals and suggests ideas. Because personal agents need a new kind of secure computer, Muse runs on Muse Secure VM, a dedicated, virtual machine (VM) that houses both the agent and a person’s data. Muse is designed around the way people already communicate, so talking to it works just like messaging another person, in the Muse app or directly in WhatsApp.

It’s simple to use. People just tell Muse what needs to get done, and it takes action, powered by Muse Spark, Meta’s most capable model to date, built for real-world agentic work like this.

Muse is designed for “non-tech” people, so anyone can use it out of the box, no technical experience needed. It handles tasks like sending an email or booking travel. Once a person shares a bigger goal, it builds a personalized plan, coordinates their time and resources, then advances the work on its own, including opening a browser, filling out forms and negotiating on their behalf.

For longer tasks, Muse keeps working after the app is closed. It returns when something changes or it needs approval, like before sending an email or making a purchase.

Checkout runs through Link by Stripe, and Muse is the first AI agent covered by Link’s purchase protections. A one-time-use card keeps real card details hidden. Shop Pay and 1Password support are coming soon.

Muse also remembers what matters to a person, like friends’ dietary restrictions before it sends invites.

Meta says Muse Secure VM brings protections no other agent offers. Muse runs on its own dedicated cloud computer, isolated from every other agent, where connected services’ data and credentials are stored.

A separate Sentinel agent on the same machine blocks anything from reaching the internet without its approval, and asks the person when needed. Muse never sees passwords or payment methods; it uses them from secure storage.

Sensitive actions, like sending an email or making a purchase, require the person’s confirmation, and a full audit trail shows what Muse has done and plans to do. People control each app’s access, can disconnect at any time, and can opt out of AI training. Conversations and VM data stay out of Meta’s ad systems, and Muse will forget anything on request.

Later this year, Muse Confidential VM will encrypt everything with a key only the person holds, so not even Meta can access it.

Meta Muse is rolling out in the US on iOS, Android, and muse.ai, and coming soon to AI glasses. It’s free for most of what people need but it will include a subscription plans for people who want to “do more.”

Anthropic Prepares Frontier Competitor to Counter OpenAI’s GPT-6 Astra

September 21, 2026

Anthropic is preparing to launch a new frontier model to counter OpenAI’s forthcoming GPT-6 Astra. The move marks a strategic shift for the safety-focused public benefit corporation, highlighting growing tensions amidst intensifying market pressures and the relentless economics of the AI ecosystem.

A Strategic Pivot in the Frontier Arms Race
Since its founding, Anthropic has centered its public messaging around cautious development, alignment research, and calls for industry-wide safety guardrails. CEO Dario Amodei recently advocated for proposals to “pace the frontier” through independent evaluators and coordinated international standards.

However, reports indicate that the company is expediting its next major release to match OpenAI’s upcoming GPT-6 Astra. Observers view this move as a striking reversal—a decision driven by the stark reality that falling behind in model capabilities risks irrelevance, regardless of a company’s safety-first philosophy. The pivot exposes the friction between adhering to measured deployment frameworks and maintaining market viability against competitors.

Doubling Down on Specialist Intelligence: Fable 5.1 and Mythos 5.1
While preparing its answer to GPT-6 Astra, Anthropic has also expanded its current product lineup with the release of Fable 5.1 and Mythos 5.1. Designed to push past general conversational interfaces, these releases target heavy enterprise usage and domain-specific problem-solving:

  • Advanced Coding & Systems Architecture: Both models deliver improved capabilities for multi-file codebase navigation, refactoring, and autonomous engineering workflows.
  • Knowledge Work & Technical Operations: Fable 5.1 and Mythos 5.1 offer elevated reasoning for dense document analysis, complex financial modeling, and structured data synthesis.
  • Scientific Research Capabilities: Anthropic highlighted that these models provide an early look at how AI systems will actively contribute to scientific discovery, offering advanced hypothesis testing and research support.

Balancing Ecosystem Growth and Safety
The push toward a new flagship model comes alongside significant operational milestones for Anthropic, including expanding enterprise adoption and advancing its scientific initiatives—such as its newly launched Bay Area biology laboratory.

As OpenAI prepares to roll out GPT-6 Astra, Anthropic’s dual strategy—releasing specialized workhorses like Fable and Mythos 5.1 while accelerating its next frontier model—demonstrates that the company intends to compete directly at the bleeding edge, even as the debate over safety guidelines continues.

Source: Reuters

Siri AI, a profoundly more capable and personal assistant powered by the next generation of Apple Intelligence, is here  

September 16, 2026

Apple;

Siri AI is a completely reimagined version of Siri that is more personal and powerful. With detailed responses and rich, natural conversations, Siri AI helps users get more done than ever before.

Built on Apple Intelligence, this new version of Siri draws on personal context understanding to help users find what they need in the moment across messages, emails, photos, and more. And with even more systemwide app actions, Siri AI can help users with tasks such as drafting an email from scratch or editing and sharing a set of photos.

This is the Siri Apple always intended it to be.

In order to use Apple Intelligence, you need to be using iOS 27, iPadOS 27, macOS 27, watchOS 27, and visionOS 27 is available on iPhone 16 models or later, iPhone 15 Pro, iPhone 15 Pro Max, iPad mini (A17 Pro), iPad models with M1 or later, MacBook Neo (A18 Pro), Mac models with M1 or later, Apple Vision Pro, Apple Watch Series 9 or later, Apple Watch Ultra 2 or later, and Apple Watch SE 3 when paired with an Apple Intelligence-enabled iPhone nearby.